What problem does it solve? AI agents that read persisted state files (like .rune/ directories) are vulnerable to indirect prompt injection, memory poisoning, and identity spoofing. This Skill validates those files before they are loaded or merged, blocking adversarial content from hijacking agent behavior. ## Core Features & Use Cases - Prompt Injection Detection: Scans for zero-width Unicode characters, hidden instruction phrases, HTML comment payloads, and suspicious base64 strings in state files. - Identity Verification via Git-Blame: Checks commit authorship on .rune/ files to detect PR poisoning from external contributors. - Content Consistency Checks: Validates decision/convention entry formats, detects slopsquatted package names, and flags security-override conventions. - Use Case: Before a session-bridge loads .rune/decisions.md into context, run this Skill to confirm the file has not been tampered with, receiving a CLEAN, SUSPICIOUS, or TAINTED verdict. ## Quick Start Run an integrity scan on all files in the .rune/ directory and report any prompt injection or tampering findings.
